In a stably subdivided population with symmetric migration, the chance
that a favoured allele will be fixed is independent of population str
ucture. However, random extinction introduces an extra component of sa
mpling drift, and reduces the probability of fixation. In this paper,
the fixation probability is calculated using the diffusion approximati
on; comparison with exact solution of the discrete model shows this to
be accurate. The key parameters are the rates of selection, migration
and extinction, scaled relative to population size (S = 4Ns, M = 4Nm,
LAMBDA = 4Nlambda); results apply to a haploid model, or to diploids
with additive selection. If new colonies derive from many demes, the f
ixation probability cannot be reduced by more than half. However, if c
olonies are initially homogeneous, fixation probability can be much re
duced. In the limit of low migration and extinction rates (M, LAMBDA m
uch less than 1), it is 2s/{1 + (LAMBDA/MS) (1 - exp (- S))}, whilst i
n the opposite limit (M, LAMBDA much greater than 1), it is 4sM/{LAMBD
A(LAMBDA + M)}. In the limit of weak selection (S much less than 1), i
t is 4sM/{(LAMBDA + 2) (LAMBDA + M)}. These factors are not the same a
s the reduction in effective population size (N(e)/N), showing that th
e effects of population structure on selected alleles cannot be unders
tood from the behaviour of neutral markers.